THE DELIVERY PATTERN

Four agents, one shared map.

Each role owns a specific decision and returns structured evidence for the next role. No agent gets to declare success on its own.

01

Planner

Converts the request and requirements artifact into an ordered plan. Calls the TAMUS AI API for reasoning, then writes a JSON handoff with requirements, dependencies, and risks.

LLM reasoningplan.json
02

Website agent

Runs a real local site checker. It tests all five pages, the custom 404 route, internal navigation, and referenced assets before producing a structured quality report.

local toolwebsite-check.json
03

Deployment agent

Runs the approved AWS deployment script in dry-run or live mode. The script keeps S3 public access blocked and configures CloudFront with Origin Access Control.

AWS tooldeployment.json
04

Verification + diagnosis

Checks the actual delivery path: HTTPS, redirects, certificate coverage, required pages, CloudFront headers, and denied direct S3 access. On failure, the LLM diagnoses from evidence.

verification tooldiagnosis.json

HANDOFFS MATTER

Every result is a file you can inspect.

The coordinator stores each step beneath a timestamped run directory. That makes the workflow auditable, repeatable, and easy to explain in a three-minute demonstration.

View the field notes →